IOP PUBLISHING

MEASUREMENT SCIENCE AND TECHNOLOGY

Meas. Sci. Technol. 24 (2013) 025301 (10pp)

doi:10.1088/0957-0233/24/2/025301

Novel gravimetric measurement

technique for quantitative volume
calibration in the sub-microliter range
Dong Liang 1,5 , Chris Steinert 2 , Stefan Bammesberger 3 , Laurent Tanguy 3 ,
Andreas Ernst 2,3 , Roland Zengerle 1,3,4 and Peter Koltay 2,3
1

HSG-IMITInstitut fur Mikro- und Informationstechnik, Georges-Koehler-Allee 103,

D-79110 Freiburg, Germany
2
BioFluidix GmbH, Georges-Koehler-Allee 103, D-79110 Freiburg, Germany
3
Laboratory for MEMS Applications, IMTEKDepartment of Microsystems Engineering,
University of Freiburg, Georges-Koehler-Allee 103, D-79110 Freiburg, Germany
4
BIOSSCentre for Biological Signalling Studies, University of Freiburg, D-79110 Freiburg, Germany
E-mail: dong.liang@hsg-imit.de

Received 18 May 2012, in final form 2 November 2012

Published 13 December 2012
Online at stacks.iop.org/MST/24/025301
Abstract
We present a novel measurement method based on the gravimetric principles adapted from the
ASTM E542 and ISO 4787 standards for quantitative volume determination in the
sub-microliter range. Such a method is particularly important for the calibration of non-contact
micro dispensers as well as other microfluidic devices. The novel method is based on the linear
regression analysis of continuously monitored gravimetric results and therefore is referred to
as gravimetric regression method (GRM). In this context, the regression analysis is necessary
to compensate the mass loss due to evaporation that is significant for very small dispensing
volumes. A full assessment of the measurement uncertainty of GRM is presented and results in
a standard measurement uncertainty around 6 nl for dosage volumes in the range from 40 nl to
1 l. The GRM has been experimentally benchmarked with a dual-dye ratiometric
photometric method (Artel Inc., Westbrook, ME, USA), which can provide traceability of
measurement to the International System of Units (SI) through reference standards maintained
by NIST. Good precision (max. CV = 2.8%) and consistency (bias around 7 nl in the volume
range from 40 to 400 nl) have been observed comparing the two methods. Based on the ASTM
and ISO standards on the one hand and the benchmark with the photometric method on the
other hand, two different approaches for establishing traceability for the GRM are discussed.
Keywords: gravimetric measurement, micro liquid-handling calibration, micro dispenser

calibration, linear regression, measurement uncertainty assessment

(Some figures may appear in colour only in the online journal)

1. Introduction

have been established over 20 years to calibrate volumetric

apparatus and liquid-handling devices with the capacity from
100 l to 2 l. Standard operation procedures (SOP), traceability
and uncertainty of the gravimetric approach in calibration of
liquid-handling devices have been researched and discussed
many times [36].
The lower limit of detection of the traditional gravimetric
measurement approach for volumetric calibration of liquid-

The gravimetric method is one of the most commonly used

quantitative and accurate volume determination methods in
research laboratories and in industry. Gravimetric volumetric
calibration standards like ASTM E542 [1] and ISO 4787 [2]
5

Author to whom any correspondence should be addressed.

0957-0233/13/025301+10$33.00

2013 IOP Publishing Ltd

Printed in the UK & the USA

Meas. Sci. Technol. 24 (2013) 025301

D Liang et al

handling systems mostly depends on the readability [7] of

the primary measuring devicethe balance. Modern precise
laboratory micro balances like XP2U/M from Mettler Toledo
Inc. (Greifensee, Switzerland) [7] provide readability down to
0.1 g that corresponds to about 0.1 nl water under standard
conditions. However, the effective measurement uncertainty
can be much larger depending on the applied measurement
method. The main reason for this is probably the evaporation
of liquid which proceeds for water at about a rate between (2
and 7) nl s1 cm2 under laboratory standard conditions (T =
20 C and P = 1013.25 hPa) according to our experiments. The
evaporation rate depends on humidity as well as further details
of the experimental setup like the balance tray, weighing dish,
the type of windshield, etc. The unavoidable loss of liquid
leads to a steady drift of the weighing mass.
Usually, every balance needs a so-called settling time of a
few seconds (usually 210 s) to reach equilibrium again after
loading the measuring target. In particular, when the liquid
is provided by an impinging droplet issued from a micro
dispensing device, a considerable momentum is transferred
to the balance weighing dish. During the settling time,
the dispensed liquid volume can lose several nanoliters by
evaporation, which in comparison to the measured volume is
no longer negligible in the sub-l range as is the case for larger
volumes. So far, no standard procedure for measurements
in the sub-l range has been established yet that takes this
effect into account. The well-known conventional gravimetric
volume measurement procedure as described, e.g., in ISO 4787
and ASTM E542which does not account for this effectis
not ideally suited for the sub-l range, but it can be used as a
guideline for adaptation of the method to smaller volumes as
described in this work.
By the use of non-contact dispensing devices, liquid
handling for laboratory automation purposes has accelerated
its tendency toward miniaturization in recent years. Several
non-contact dispensers for the sub-l range like PipeJet
(BioFluidix GmbH, Freiburg, Germany) [8], Multidrop
Microplate Dispensers (Thermo Fisher Scientific Inc.,
TM
Waltham, MA, USA) [9] and Deerac GX Series (Labcyte
Inc., Sunnyvale, CA, USA) [10] present very good
precision according to relative volume measurements. Such
measurements in the sub-l range are usually based on
optical and photometric methods [11, 12] that are partly also
routinely applied in the calibration of diagnostic laboratory
automation systems. However, these methods are usually
not able to provide traceable volume results and a reliable
uncertainty analysis regarding all referenced input quantities
in the measurement chain. The only exception until now is
the commercial liquid measurement system from Artel Inc.
(Westbrook, ME, USA)the multichannel verification system
(MVS) [13, 14]. This system is based on a dual-dye ratiometric
photometric method which provides traceability according to
the National Institute for Standards and Technology (NIST)
[15] in the volume range down to 30 nl. However, this method is
quite complex and limited to the use of a specific testing liquid
(proprietary dyewater mixture from Artel Inc.). For liquids
other than water, a separate calibration of the photometric setup
is required and the traceability is lost.

In comparison to photometric measurement principles,

the gravimetric measurement is largely independent of the
liquid used for measurement, which is an important feature for
application near characterization of dispensing devices. Only
the effective density has to be known to relate the measured
mass to the dispensed volume. Environmental influence factors
like temperature or pressure can be considered straightforward
with the same corrections as for larger volume measurements.
Therefore, the gravimetric method should in principle be able
to provide similar benefits to micro volume characterization.
In previous work it has been proposed to combine
photometric measurements with gravimetric determination
of the total mass of the well plate to determine the
average accuracy [12, 16]. Although this method provides an
improvement compared to single-color photometric methods,
the absolute volume of a single dispense cannot be
determined gravimetrically. Therefore, the direct gravimetric
measurement of a liquid aliquot as recently described in
[17] would be more desirable. In this publication, the
authors demonstrate the basic feasibility of gravimetric liquid
measurements in the sub-l range considering evaporation,
but they fall short of providing a proper uncertainty analysis
and do not establish traceability of the proposed measurement
approach according to international standards. In particular,
the effect of non-equilibrium balance oscillations and the error
of the linear regression calculation are not considered there.
Both of which can be the major sources of error. In order to
overcome these limitations, the gravimetric regression method
(GRM) as described in the following and two different ways
to establish traceability for this method are presented in this
work.

2. Description of the GRM

In order to determine the volume of a liquid aliquot in the
sub-l range based on the gravimetrically measured mass
change caused by this liquid volume on a high precision
balance, similar equations as provided by the ASTM E542
and ISO 4787 standards are used. The formulae and the basic
measurement procedure are discussed in other publications for
large volume calibrations in detail [16]. For the purpose of
this work, the equation for determination of the liquid volume
V20 based on a gravimetrically measured mass m is considered
as follows:


1
a
,
(1)
V (T ) = m
1
w (T ) a
c
V20 = V (T ) [1 (T 20)],

(2)

where V (T ) is the total volume of measured liquid contained

in the weighing dish at measurement temperature T . The
temperature of measured liquid is supposed to be the same
as air temperature during measurement. V20 is the volume at
reference temperature 20 C and m is the mass of the liquid
contained in the weighing dish on the balance. w is the density
of the measured liquid at measurement temperature T . a is
the density of air, approximately to be 0.0012 mg l1. c
is the density of the weights used to calibrate the balance
2

Meas. Sci. Technol. 24 (2013) 025301

D Liang et al

(8.006 103 mg l1 according to the product certificate of

the balance used in this work). is the thermal coefficient of
expansion of the measured liquid.
The measured primary quantity is the total mass m on
the weighing dish of the balance, which is conventionally
weighed by the used balance. The mass change caused by
the addition of a liquid volume V is denoted by m. The
mass of the additional liquid is supposed to be the difference
of mass before (mbefore ) and after (mafter ) the dispensing event
as given in equation (3):
m = mafter mbefore .

(3)
Figure 1. Plot of a typical measurement: balance readouts (dots)
before and after dispensing are displayed as a function of time.
Readouts acquired between the two vertical lines are not used for
calculation (settling time of the balance).

The mass values mbefore and mafter are of course subject to a

measurement error m resulting from the error of the weighing
process especially from the error of the used gravimetric
balance.
Simply speaking the mass of the dispensed liquid volume
is mainly determined by the difference of the mass before
and after the dispensing event. Usually, the average values
of a series of weighing results that are acquired before the
dispensing event and after a settling time of a few seconds are
considered as appropriate measurement results for mbefore and
mafter . However, due to evaporation, the mass of the liquid
loaded on the weighing dish of the balance changes over
time. After the settling time, the mass detected is no longer
corresponding exactly to the mass directly after the dispensing
event. It is therefore not admissible to consider simple average
values of balance readouts for determining mbefore and mafter .
One possible method for overcoming this problem is the
linear regression method described in [17]. In this method,
the value determined by a linear regression based on several
balance readouts acquired before and after the dispensing
event is regarded as an appropriate measure for the mass
before and after the dispensing event. However, as liquid keeps
evaporating from the weighing dish, the balance never reaches
equilibrium during the measurement, and thus, the balance
reading does not exactly correspond to the real mass value.
It is therefore questionable to consider the balance readouts
under such non-equilibrium conditions as mass equivalents to
determine the values mbefore and mafter without any further
consideration of the validity of the results. Therefore, the
uncertainty of the linear regression analysis has to be taken
into account as an indicator of the validity and reliability of
the balance readings.
In order to establish an uncertainty analysis for the
GRM, the gravimetric measurement procedure is considered
as follows. A weighing dish filled with evaporating liquid
measured repeatedly n times results in a series of decreasing
balance readouts Ii at time ti. The ith weighing readout can
then be supposed to yield the mass mi of the weighing
dish filled with liquid and an independent error i which
represents a random measurement error. Assuming that the
evaporation rate of the testing liquid is constant under ideal
laboratory conditions with constant temperature, humidity and
air pressure, mi is supposed to have a linear relation to time ti .
And the balance readings Ii can be expressed as follows:
Ii = mi + i = a bti + i .

Assume that each i obeys an independent distribution with

variance of i2 ; then, with the method of least squares,
a and b can be estimated based on a weighing series as
follows [17]:
n ti n ti Ii
n ti2 n Ii
i=1 i2
i=1 i2
i=1 i2
i=1 i2
;
a=

n 1 n ti2
n ti 2
i=1 2
i=1 2
i=1 2
i

n

b=

1
i=1 i2

n

n

1
i=1 i2

ti Ii
i=1 i2

n

ti2
i=1 i2

n

ti
i=1 i2


n

n

ti
i=1 i2

Ii
i=1 i2
2

(5)

Here, n is the total number of weighings and i2 is the variance

of i .
Under certain conditionsexplained in more detail
belowit can be assumed that i2 = 2 = const, and the
following simplified equations can be applied to calculate a
and b:
n
n
n 2 n
i=1 ti
i=1 Ii
i=1 ti
i=1 ti Ii
;
a=
n 2 n 2
n i=1 ti
i=1 ti



n ni=1 ti Ii ni=1 ti ni=1 Ii
.
(6)
b=
n 2

n ni=1 ti2
t
i
i=1
If there were no evaporationas is the case for solid objects
then the balance readouts Ii would oscillate around the
mean mass value with decreasing amplitude after loading
of the weighing dish until the balance mechanism reached
equilibrium again. After a settling time, Ii can then be
supposed to be equal to mi , which means that the expectation
value of i is zero. However, in the case of liquid, the
load is continuously changing and the balance never reaches
equilibrium. As a consequence, the mass readouts Ii are
continuously oscillating around the actual equilibrium value
with is constantly decreasing, as shown in figure 1. It has to
be pointed out that these non-decaying periodic oscillations are
not caused by the droplet impact onto the weighing dish, but
appear as a consequence of the (electro-) mechanical properties
of the balance mechanism and do not decay until all the liquid
on the weighing dish has completely evaporated.

(4)
3

Meas. Sci. Technol. 24 (2013) 025301

D Liang et al

of the mass value to the point tdisp in time. Since both

mLR,before (tdisp ) and mLR,before (tdisp ) are time-projected values,
it is mandatory to consider the uncertainties coming from the
prognostic calculation which are denoted as u(mLR,before (tdisp ))
and u(mLR,after (tdisp )). Considering u(mLR,before (tdisp )) and
u(mLR,after (tdisp )) correctly makes a major difference in the
uncertainty analysis of the measurement result when compared
to the method described in [17].
For determining the uncertainty u(mLR,before (tdisp )) and
u(mLR,after (tdisp )) stemming from the regression analysis, the
prognosis interval (mLR (t ) Dprog (t )) has been applied. This
interval covers the expectation value under a probability of
(1 ) at time t and the half-width of the prognosis interval
Dprog (t ) at time t is calculated as follows [18, 19]:

With constant evaporation rate b and constant time

intervals ti ti1 between the readouts, the change of mass
due to evaporation with respect to the previous measurement
mi mi1 can be supposed to be the same at each time
ti . Therefore, the approximation of i2 by a constant value
is justified and the regression line provides a reasonable
estimate of the continuously decreasing mass value. With
equation (6) it is easy to obtain the linear regression parameters
(abefore , bbefore ) and (aafter , bafter ) independently, based on two
series of balance readouts before and after dispensing. Using
the parameters (abefore , bbefore ) and (aafter , bafter ), two values
mLR,before (tdisp ) and mLR,after (tdisp ) can be readily calculated,
which are the projected mass values at the time of dispensing
tdisp :
mLR,after (tdisp ) = aafter + bafter tdisp ,
mLR,before (tdisp ) = abefore + bbefore tdisp .

(7)

Dprog (t ) = tn2,1 2 s

Here, tdisp indicates the moment when the dispense of the liquid
volume is triggered, which isexcept for an offset of a few
millisecondsthe point in time when the liquid just impinges
on the weighing dish.
Denoting the values estimated by the linear regression by
subscripts LR, m is supposed to be the difference of these
two linear regression values at time tdisp :
m = mLR,after (tdisp ) mLR,before (tdisp ).

1+

(t t )2
1
+ n
,
2
n
i=1 (ti t )

(9)

where


n
 1
(mi mLR,i )2 ,
s=
n 2 i=1

(8)
1
ti .
n i=1
n

The analysis of the regression coefficients before and after

dispense enables one to verify the validity of the measurement
result m obtained this way. If, e.g., bbefore and bafter deviate
significantly from each other, this indicates unsteady and
therefore unreliable measurement conditions. Otherwise, the
evaporation rates would be constant, and bbefore and bafter would
be the same. Such unreliable conditions might be caused by
environmental effects like vibrations of the balance due to wind
gust or mechanical disturbance. Therefore, the comparison
of the regression parameters bbefore and bafter provides an
important means to assess the quality and reliability of
the measurement, by verification of the constancy of the
evaporation rate.
For the experiments reported in this work the
measurement conditions have been established that consistent
regression parameters before and after dispensing with less
than 5% variation could be obtained. The data acquisition
time before and after the measurement has been set to 16 s
corresponding in our case to 32 data points to obtain a
sufficiently detailed statistic on the regression analysis and
the uncertainty estimation. Further details of the experimental
setup that were applied to reduce statistical errors are given in
section 4.

t=

Here, s is the deviation of the linear regression residuals

for least squares calculation, t is the average time of n
measurement points, mLR,i is the resulting mass determined
from the linear regression at time ti , n is the total number
of weightings and tn2,1 2 is the t-distribution (Student
distribution)
 for (n 2) degrees of freedom with a probability

of 1 2 [18, 19]. Finally, is the probability of error (taken
as 5% for all calculations presented here).
In the case of = 0.05, the prognosis interval
(mLR (tdisp ) Dprog (tdisp )) covers with 95% probability the
value of the measured mass at time tdisp , corresponding to
the expanded uncertainty in the Guide to the expression
of uncertainty in measurement (GUM) [20] with k = 2.
Here, k denotes the kth -quantile of the Gaussian normal
distribution. Thus, the uncertainties u(mLR,before (tdisp )) and
u(mLR,after (tdisp )) can be expressed as follows:
u(mLR,before (tdisp )) = 12 Dprog,before (tdisp ),
u(mLR,after (tdisp )) = 12 Dprog,after (tdisp ).

3. Uncertainty estimation of the GRM

(10)

Based on this uncertainty assessment, the combined standard

uncertainty u(V20 ) of the measured volume V20 , calculated
according to equations (1) and (2) using the mass difference
determined from the regression, can be expressed according to
EURAMET cg-19 [6] and the method described in the GUM
[20] as follows:

As explained before, the balance never reaches equilibrium

during measurement due to the mass change caused by
evaporation. Although the linear regression method accounts
for the evaporation, it also induces an error into the calculation
of the measurement result m due to the back projection
4

Meas. Sci. Technol. 24 (2013) 025301

D Liang et al

Table 1. Uncertainty calculation of GRM for a typical measurement (the same measurement data as shown in figure 1).
Input quantity ( )

Unit

Value

Standard uncertainty u( )

m
m
w
a
c

g
g
g nl1
g nl1
g nl1
1
C

199.0
0
0.996 6465
0.0012
8.006
0.000 207
26.7

5.9842
3.05
1.52E-06
2.89E-07
0.01
2.89E-07
0.05

V20
u(V20)
U(V20), k=2

nl
nl
nl

199.6
6.7
13.5




V20 2 2
V20 2 2
u(V20 ) =
u (m) +
u (m)
m
m


2
2
V20
V20
u2 (w ) +
u2 (a )
+
w
a




V20 2 2
V20 2 2
u (c ) +
u ( )
+
c



V20 2 2
u (T ),
+
T
where


Sensitivity coefficient
1.0030
1.0030
200.5053
172.3952
0.0037
1339.1259
0.0414

even larger than u(m) for the presented case, it is evident how
important it is to consider the error from the regression analysis
and the prognostic calculation correctly, in order to avoid
underestimation of the uncertainty of the final measurement
result.
To enable the direct comparison with the Artel MVSfor
which the uncertainty is provided by the manufacturer for the
95% confidence interval (i.e. expanded uncertainty, k = 2)
the uncertainty for the GRM will also be discussed in terms
of the expanded uncertainty which can be calculated from
the standard uncertainty u(V20 ) by multiplication with k = 2:
U (V20 ) = 2 u(V20 ).

(11)

u2 (m) = u2 (mLR,after (tdisp )) + u2 (mLR,before (tdisp )). (12)

The value of u(m) in equation (11) is given by the
uncertainty obtained from the Swiss Calibration Service
(SCS) for the Mettler-Toledo XP2U/M balance used for
the presented experiments (calibration identification number
140538). The SCS calibration is based on OIML R1111:2004
[21]. According to the calibration certificate, the expanded
uncertainty (k = 2) of weighing conventional mass (stainless
steel with the density of 7950 140 kg m3) of 100 mg
is 2.6 g and of 2 g is 6.1 g. This uncertainty value can
be assumed as the uncertainty u(m) of a normal weighing
process including the error of repeatability, linearity, eccentric
loading and sensitivity offset of the gravimetric balance. The
value u(m) stems from the uncertainty analysis of the linear
regression calculation presented before. It is calculated based
on the balance readouts according to equation (12). The
uncertainty of the other factors in equation (11) concerns
the volume conversion from mass as well as temperature
corrections, which are calculated in a similar way as performed
by the well-known standard procedures for macroscopic
volume measurements [1, 6].
The values of the individual uncertainty components and
the finally resulting measurement uncertainty for a typical
measurement according to the proposed method based on the
measurement results shown in figure 1 are given in table 1. The
total standard measurement uncertainty for measuring a liquid
volume of about 200 nl results in u(V20 ) = 6.7 nl. In contrast to
u(m), which represents the uncertainty of a normal weighing
process of a non-evaporating mass, the uncertainty u(m)
takes the regression calculation and the balance oscillation into
account. Therefore, both of these values have to be considered
for calculating the resulting uncertainty u(V20 ) of the final
volumetric result V20. Since u(m) has considerable size,

4. Experimental section
The primary measurement device of the gravimetric
liquid measurement system used in this study is a
high precision laboratory micro balance type XP2U/M
from Mettler Toledo, Switzerland. A micro weighing dish
(ID H 13 mm 3.5 mm, VWR Order no 6111356)
filled with about 200 l of the same testing liquid that is
provided to the dispensing device (i.e. deionized water or
Artels proprietary dyewater mixture) is placed at the center
of the weighing pan and is covered by an aluminum draft
shield (ID H 80 mm 24 mm) with a circular opening
( 6 mm) on the top. The testing liquids involved in this
experiment are deionized water with water quality of 18 M

cm (Quality 1 in ISO 3696) [22, 23] and Artel sample solution

(Artels proprietary dyewater mixture) of type C (for volumes
from 500 nl to 2.5 l in Artel standard 384-well plates,
density 1.0059 g nl1 at room temperature according to
manufacturer), type D (for volumes from 150 nl to 500 nl
in Artel standard 384-well plates, density 1.0020 g nl1 at
room temperature according to manufacturer) or type E
(for volumes from 30 to 150 nl in Artel standard 384-well
plates, density 1.0077 g nl1 at room temperature according
to manufacturer). For temperature correction to reference
temperature 20 C, we used the same thermal coefficient of
expansion for all types of liquids. The nozzle of the dispensing
device to be characterized is put directly in closest proximity
over the opening of the draft shield. The balance as well as the
dispensing device are placed on a vibration isolated granite
table (Johann Fischer Aschaffenburg DIN876126 hardstone
surface plate and Newport VH3660W-OPT vibration isolated
workstation) to reduce mechanical vibrations and a large
5

Meas. Sci. Technol. 24 (2013) 025301

D Liang et al

Figure 2. Experimental setup for gravimetric measurement as well as for micro plate preparation for subsequent photometric evaluation
(equipment of Artel-MVS not shown). Di-H2O: laboratory deionized water.

above. Directly after the gravimetric measurement the PipeJet

was automatically moved to fill the same volume into 21
wells of a certified Artel 384-well micro plate (that had been
prepared before, according to the method described in detail
in [13, 14]) for subsequent spectrophotometric measurement.
The first dispense of a measurement series was always used
as priming shot and subsequently excluded from evaluation.
Thus, the number of results reduces to n = 20 per experiment
in both cases. Such a priming shot is important to reduce the
effect of liquid evaporation from the nozzle after a certain
undetermined idle time of the dispenser that can lead to a
drift of the droplet volume or dye concentration which causes
systematic errors.
After one of these measurement cycles, the PipeJet moved
back to the balance and started the next measurement cycle
with a different dispensing volume. Different dispensing
volumes were generated by adding multiple droplets (1, 2, 4,
etc) to the balance weighing dish. In total, eight different test
volumes from about 40 nl (i.e. one droplet) to about 1 l (i.e.
25 droplets) were measured, whereas the PipeJet dispenser
was always operated by the same parameters (down-stroke
length = 25 m, up-stroke velocity = 1 m ms1, hold time =
10 s, down-stroke velocity = 100 m ms1 and shot delay =
50 ms) to produce an individual droplet. Details of the PipeJet
working principle and the meaning of the individual control
parameters are described in [8].
After the completion of three consecutive experiments
with three different volumes, the automatic execution of
the experiment was stopped and the filled micro titer plate
was processed according to the Artel MVS method. This
interruption was necessary because the accuracy of the Artel

wind shield cover made of transparent PMMA is placed over

the whole measurement setup to isolate it from surrounding
air convection as shown in figure 2. The acquisition of the
balance readouts, the evaporation compensation according to
the proposed linear regression method and the calculation of
uncertainty of GRM as well as triggering the dispenser were
automatically executed by a self-programmed vb.net software
program.
In order to investigate the performance of the proposed
gravimetric volume measurement procedure and experimental
setup, a dispensing device with high repeatability and
reproducibility is required. Therefore, a PipeJet P9 dispenser
(BioFluidix GmbH, Freiburg, Germany) with 500 m
dispensing pipe was selected. This dispenser is known to
operate with very high reproducibility (CV: coefficient of
variation <1%) at single droplet volumes in the range from
10 to 60 nl. The PipeJet P9 dispenser was mounted on a
liquid-handling robot (BioSpot, BioFluidix GmbH, Freiburg,
Germany) in such a way that liquid could be dispensed onto the
micro balance or into an Artel standard 384-well plate. Thus,
the identical setup was used for gravimetrical measurements
as well as to fill liquid into a micro well plate for executing the
certified measurement method with the Artel MVS system.
For a first benchmark between the Artel MVS method and
the proposed gravimetric method, the balance weighing dish
and the PipeJet P9 dispenser were loaded with Artel sample
solution (proprietary dyewater mixture from Artel Inc.). Artel
sample solution of type C, D or E was used for the different
volume ranges as described in the operating instructions. First,
the PipeJet was automatically moved over the micro balance
to execute 21 times the gravimetric measurement as described
6

Meas. Sci. Technol. 24 (2013) 025301

D Liang et al

of type C, D or E are the same. The time interval between

these two experimental series with the same hardware and
dispensing parameters was about 12 h with an ambient
temperature difference of about 3 C.

5. Results and discussion

The benchmark results of the gravimetric measurements and
the photometric Artel MVS measurements are presented in
figures 3 and 4. The repeatability given by the coefficient of
variation (CV) is below 3% for both measurement methods
in the whole considered measurement range, where the CV
is defined as the ratio of the standard deviation to the mean
measured volume. Furthermore, both measurement methods
provide greatly consistent results. The small deviations
between the results of the two different methods are best
discussed in terms of the so-called bias which can be expressed
in terms of the absolute bias (nl) and relative bias (%) defined
as follows:
Bias [nl] = GRM.V20 Artel.V20,
(13)
Bias [%] = GRM.V20Artel.V20
100%.
Artel.V20

Figure 3. Benchmark results of the presented gravimetric

measurement method compared to the Artel-MVS method. Standard
deviation of 20 measurements is shown as error bars in the x- and
the y-direction. Environmental conditions during the measurements:
temperature = 29.4 C, relative humidity = 51%, air pressure =
975 hPa. GRM: gravimetric regression method.

method is also influenced by evaporation effects, such that

prepared well plates should not be stored for an extended
period of time before reading. Furthermore, liquids C, D or E
had to be supplied to the dispenser when the corresponding
volume range was addressed. After reading of the microliter
plate results with the Artel MVS reader, a new plate was
prepared and the experiments have been continued as described
above.
After the benchmark experiment with the Artel sample
solution of type C, D or E, the gravimetric measurement
was carried out once more with the same experimental setup
but using deionized water. The same dispensing parameters,
as well as measurement configuration was used like before,
except of the fact that dispensing was performed onto
the balance only and no Artel MVS measurements were
made this time. This second experiment was carried out in
order to determine whether the measurement results obtained
gravimetrically with deionized water and the Artel solution

Here, GRM.V20 is the measured mean volume at the

reference temperature 20 C determined with the presented
GRM approach and Artel.V20 is the measured mean volume
at reference temperature 20 C using the Artel MVS with
temperature correction according to equation (2).
The bias represents the difference between the mean
volume measured by the GRM approach and the reference
value given by the Artel MVS. In the lower volume range from
40 to 400 nl, the absolute bias is almost constant around 7 nl.
For larger volumes, the absolute bias becomes somewhat
larger and reaches up to 27 nl. Given the measurement
uncertainty of the gravimetric method and the Artel method
[24] as shown in figure 4, all measured values have to be
considered to be consistent, since the bias is always smaller
than the corresponding measurement uncertainty.
In view of the proven consistency of both methods,
one could argue that traceability of the gravimetric method
could be established by calibration of the gravimetric results

Figure 4. Bias and CV of the presented gravimetric measurement method compared to the Artel-MVS method. Environmental conditions
during the measurements: temperature = 29.4 C, relative humidity = 51%, air pressure = 975 hPa. GRM: gravimetric regression method.
7

Meas. Sci. Technol. 24 (2013) 025301

D Liang et al

not with deionized water as requested in ASTM E542 and ISO

4787.
Assuming though that the ASTM E542 and ISO 4787
could be extended to the microscopic volume range by
the described GRM method, only the consistency of using
Artel solution compared to deionized water for gravimetric
measurement has to be shown to provide direct traceability of
the gravimetric method. In order to verify whether the usage
of the Artel sample solution of type C, D, or E gives the
same gravimetric measurement results compared to dispensing
deionized water, the second experimental series was performed
as described above using deionized water only. As shown in
figures 5 and 6, the same results have been obtained by the
gravimetric method using either the Artel solution of type C,
D, or E or deionized water in the whole volume range. The
volumes obtained for either liquid are consistent within a few nl
only (cf figure 6). This indicates that first the PipeJet dispenser
delivers the same volumes with either liquid and, second, the
gravimetric results are not influenced by the choice of either
liquid. Both of which could be expected due to the very similar
density and evaporation behavior of both liquids. Therefore,
it seems admissible to use the measured data obtained for
deionized water as well as for the Artel solution of type C, D,
or E to establish traceability for the proposed method according
to ASTM E542 and ISO 4787 (provided that the standard can
be considered as extendable to small volumes down to 40 nl
when the described GRM approach is used). Thus, traceability
for the GRM could be established consistently by two different,
independent approaches.

Figure 5. Comparison of gravimetric measurements carried out with

Di-H2O and Artel sample solution of type C, D or E. Standard
deviation of 20 measurements is shown as error bars in the x- and
the y-direction. Environmental conditions for measurement with
Artel sample solution: temperature = 29.4 C, relative humidity =
51%, air pressure = 975 hPa. Environmental conditions for
measurement with Di-H2O: temperature = 26.7 C, relative
humidity = 52%, air pressure = 975 hPa. GRM: gravimetric
regression method. Di-H2O: laboratory deionized water.

through the benchmark with the traceable Artel measurements.

This calibration line would be indeed given by figure 3. On
the other hand, traceability is better established by relating
the gravimetric volume measurement results directly to the
gravimetric mass change acquired by the balance analogue to
ASTM E542 and ISO 4787. However, ASTM E542 and ISO
4787 are not strictly valid for the small volumes considered
herein particular due to the discussed significant evaporation
effectand a corresponding standard for the sub-l range is
not yet available. Furthermore, the gravimetric results have
been obtained by using Artel sample solutions C, D, E and

6. Conclusion
In this paper, we have presented an improved gravimetric
volume measurement method referred to as gravimetric
regression method (GRM) for determining liquid volumes in
the sub-l range. The core element of this method is a linear
regression for evaporation compensation applied on a large
number of measured data points before and after addition
of the liquid volume to the balance. In order to predict the

Figure 6. Bias and CV of the presented gravimetric measurement method carried out with Di-H2O and Artel sample solution.
Environmental conditions for measurement with Artel sample solution: temperature = 29.4 C, relative humidity = 51%, air pressure =
975 hPa. Environmental conditions for measurement with Di-H2O: temperature = 26.7 C, relative humidity = 52%, air pressure =
975 hPa. GRM: gravimetric regression method. Di-H2O: laboratory deionized water.
8

Meas. Sci. Technol. 24 (2013) 025301

D Liang et al

larger uncertainties for such disturbed measurements. Thus, in

summary, the presented GRM approach is suitable to provide
flexible, accurate, objective and traceable results for liquid
volume measurements in the sub-l range. This will be very
helpful for quantitative calibration of micro non-contact liquidhandling devices and other microfluidic applications in the
future.

mass difference caused by the dispensed liquid correctly, the

prognostic error of the linear regression has to be considered.
Therefore, a thorough uncertainty analysis has been presented
for the GRM and a combined standard uncertainty u(V20 ) of
typically 6 nl and an expanded uncertainty U (V20 ) of typically
12 nl have been determined for volumes in the range from
40 nl to 1 l.
According to the presented benchmark experiment, the
results of the gravimetric measurements have been shown to
be consistent with the Artel MVS photometric method that was
used as a reference. To the knowledge of the authors, this is the
first time that traceability according to international standards
has been established for a volumetric measurement carried out
in the sub-l range by means of a gravimetric method. The
major advantage of such a gravimetric method compared to
photometric approaches is that it can be applied in principle
with any liquid which makes it more flexible and easier to use
in application near measurement scenarios.
The analysis of the measurement errors revealed that the
uncertainty of the new GRM is larger than the uncertainty of
the photometric reference method (Artel MVS) in the lower
volume range but smaller for volumes larger than 700 nl.
However, since the central measuring device in the GRM, i.e.
the balance, normally provides a smaller uncertainty than the
spectrophotometer in the Artel MVS photometric method, the
GRM may have the potential to deliver a smaller uncertainty if
the measurement procedure is further improved. In this second
approach, traceability to international standards is established
independently of the Artel MVS photometric method, but
relying on the traceability of the balance itself, analogous to
the standards ASTM E 542 and ISO 4787 for large volume
measurements. Although these standards are not strictly valid
for the considered volume range, they still can be considered
applicable for the purpose of this work in view of the fact
that more suitable standards for the sub- l range do not exist
yet. Furthermore, the experimental data obtained in this study
prove that both methods to establish traceability yield the
same results within the limits of uncertainty, which justifies
the proposed approach for direct calibration ex post.
Compared to the previous work [17], this study has
provided a thorough uncertainty analysis of the GRM and has
established traceability according to international standards for
this method. It turned out that the uncertainty caused by the
use of the prognostic value of the linear regression performed
on the primary balance readouts is significant and must not
be underestimated. Furthermore, it has been found that the
fitted value for the rate of evaporation and the magnitude of
the prognosis interval can be used as criteria to determine
the reliability of a measurement. If the fit values for the
periods before and after addition of the volume to the balance
deviate too much from each other and/or the prognosis interval
becomes large, this indicates a disturbance of the measurement
setup during data acquisition. If such a disturbance is
considered to be too large, the corresponding measurement
can be rejected. However, if the statistical measurement
error caused by the disturbance is considered to be still
in an acceptable range, the presented uncertainty analysis
accounts for this error properly and delivers correspondingly

Acknowledgment
The authors gratefully acknowledge financial support from the
German Federal Ministry for Science and Education (BMBF)
through the project Smart Reagent Dosage (SFK 16SV5119).
Also, we would like to thank our project partners from
Roche Diagnostics GmbH, Mannheim, Germany, for their kind
support and valuable feedback.

References
[1] ASTM 2007 Standard practice for calibration of laboratory
volumetric apparatus ASTM E542-01
[2] DIN ISO 2011 Laboratory glasswarevolumetric
instrumentsmethods for testing of capacity and for use
DIN EN ISO 4787: 2011-05
[3] Batista E, Pinto L, Filipe E and Vanderveen A 2007
Calibration of micropipettes: test methods and uncertainty
analysis Measurement 40 33842
[4] Lorefice S 2009 Traceability and uncertainty analysis in
volume measurements Measurement 42 15105
[5] DIN ISO 2007 Piston-operated volumetric apparatus: part 6.
Gravimetric methods for the determination of measurement
error DIN EN ISO 86556: 2002-12
[6] EURAMET 2011 Guidelines on the determination of
uncertainty in gravimetric volume calibration
EURAMET/cg-19/v2.1 http://www.euromet.org/
fileadmin/docs/Publications/calguides/EURAMET-cg19__v_2.1_Guidelines_in_uncertainty_volume.pdf
[7] XP2U Ultra-Microbalance, METTLER TOLEDO, Greifensee,
Switzerland http://de.mt.com/de/en/home/products/
Laboratory_Weighing_Solutions/
MX-UMX/XP2U_Ultra-microbalance_1.html (last
accessed 14 November 2011)
[8] Streule W, Lindemann T, Birkle G, Zengerle R and Koltay P
2004 PipeJet: a simple disposable dispenser for the nanoand microliter range J. Assoc. Lab. Autom. 9 3006
[9] Multidrop Microplate Dispensers, Thermo Fisher Scientific,
Waltham, MA, USA
http://www.thermo.com/com/cda/landingpage/
0,10255,764,00.html?ca = multidrop (last accessed 14
November 2011)
[10] DeeracTM gx Series, Labcyte Inc., Sunnyvale, CA, USA
http://www.labcyte.com/Deerac TM_GX_Series/
Default.193.html (last accessed 14 November 2011)
[11] Thurow K, Kruger T and Stoll N 2009 An optical approach for
the determination of droplet volumes in nanodispensing
J. Autom. Methods Manage. Chem. 2009 198732
[12] Taylor P B et al 2002 A standard operating procedure for
assessing liquid handler performance in high-throughput
screening J. Biomol. Screening 7 55469
[13] MVS Multichannel Verification System for Automated Liquid
Handlers and Multichannel Pipettes, Artel, Westbrook, ME,
USA http://www.artel-usa.com/products/
mvs_advanced.aspx (last accessed 14 November 2011)
[14] Bradshaw J, Knaide T, Rogers A and Curtis R 2005
Multichannel verification system (MVS): a dual-dye
9

Meas. Sci. Technol. 24 (2013) 025301

[15]

[16]

[17]

[18]
[19]

D Liang et al

ratiometric photometry system for performance verification

of multichannel liquid delivery devices J. Assoc. Lab.
Autom. 10 3542
Artel Multichannel Verification System Traceability Document
24A5025G http://www.artel-usa.com/uploadedFiles/
Controlled_Documents/MVS_Controlled_Documents/
MVS Traceability Document 24A5025.pdf (last accessed
22 October 2012)
Rhode H, Schulze M, Renard S, Zimmermann P, Moore T,
Cumme G A and Horn A 2004 An improved method for
checking HTS/uHTS liquid-handling systems J. Biomol.
Screening 9 72633
Verkouteren R M and Verkouteren J R 2009 Inkjet metrology:
high-accuracy mass measurements of microdroplets
produced by a drop-on-demand dispenser Anal. Chem.
81 857784
Press W H, Teukolsky S A, Vetterling W T and Flannery B P
2002 Numerical Recipes in C: The Art of Scientific
Computing (Cambridge: Cambridge University Press)
Ross S M 2004 Introduction to Probability and Statistics for
Engineers and Scientists (Amsterdam: Elsevier)

[20] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML
2008 Guide to the Expression of Uncertainty in
Measurement JCGM 100:2008 GUM 1995 with minor
corrections www.bipm.org/utils/common/documents/jcgm/
JCGM_100_2008_E.pdf
[21] OIML 2004 International Recommendation Weights of Classes
E1, E2, F1, F2, M1, M12, M2, M23 and M3 Part I:
Metrological and Technical Requirements OIML R 111-1
edn 2004(E) www.oiml.org/publications/R/R111-1-e04.pdf
[22] IMTEK Reinraumlaborkurs 1 SS2009 Skript 2011
http://www.imtek.de/anwendungen/content/upload/
vorlesung/2009/ss2009_skript_reinraumlaborkurs1_1.pdf.
(last accessed 14 November 2011)
[23] DIN ISO 1991 Water for analytical laboratory
usespecification and test methods DIN ISO 3696:
1991-06
[24] Artel MVS Uncertainty of measurement Document
24A5026D http://www.artel-usa.com/uploadedFiles/
Controlled_Documents/MVS_Controlled_Documents/
24A5026D_MVS Uncertainty of Measurement.pdf (last
accessed 14 November 2011)

10